[ArXiv] 2nd week, Oct. 2007

Frankly, there was no astrostatistically interesting paper from astro-ph this week but profitable papers from the statistics side were posted. For the list, click

In terms of detecting sources, my personal thought is that the topics of size, power, false discovery rates, and multiple testing ask for astronomers attention. To my understanding, the power of observation has increased but the methodology of source detection cites tools of two to three decades ago when assuming normality, iid, or homogeneity for data and instruments was agreeable.

3 Comments
  1. vlk:

    “the methodology of source detection cites tools of two to three decades ago when assuming normality, iid, or homogeneity for data and instruments was agreeable”

    What methodology are you referring to? The state of the art source detection tool for X-ray data is wavdetect, which assumes neither normality nor homogeneity.

    10-12-2007, 5:19 pm
  2. hlee:

    I was away from slogging for a week. My implication was any kind of thresholding to declare sources. It seems that wavdetect is a wavelet based method, which I believe transforms from image space to wavelet space and raises a red flag when a wavelet coefficient is high enough (I’ll check the method carefully later to make sure what statistics involved in there. Sadly, my illiteracy to wavelets will delay it and the following could be irrelevant).

    I’d rather address questions before defending myself. Does this wavdetect or other source detection methods recovers sources 100%? What is the power of these methods? To my understanding, source detection methods are designed to detect multiple sources for cataloguing before labor intensive human eye verification. Power varies depending on methods to the same data set in addition to type I errors. When it comes to multiple sources, a simple thresholding looses power and one just cannot lower the threshold to increase power because type I error increases. More careful multiple testing schemes are required to say that source detecting methods recover sources with a certain confidence.

    10-19-2007, 12:47 pm
  3. vlk:

    Nobody says wavdetect is perfect. See here:
    http://hea-www.harvard.edu/AstroStat/slog/2007/the-power-of-wavdetect/

    10-21-2007, 4:01 pm
Leave a comment